Speech recognition method and speech recognition apparatus

• US 6,067,513 A
• Filed: 10/22/1998
• Issued: 05/23/2000
• Est. Priority Date: 10/23/1997
• Status: Expired due to Fees

• Current Assignee: Pioneer Electronics (USA) Incorporated (Pioneer Corporation)
• Original Assignee: Pioneer Electronics (USA) Incorporated (Pioneer Corporation)
• Inventors: Ishimitsu, Shunsuke
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): ABEBE, DANIEL DEMELASH
• Application Number:
  US09/176,302
• Time in Patent Office: 579 Days
• Field of Search: 704/233, 704/226, 704/256
• US Class Current: 704/233
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 21/0216 characterised by the method used for estimating noise
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum

Abstract

A speech recognition method of recognizing an input speech in a noisy environment by using a plurality of clean speech models is provided. Each of the clean speech models has a clean speech feature parameter S representing a cepstrum parameter of a clean speech thereof. The speech recognition method has the processes of: detecting a noise feature parameter N representing a cepstrum parameter of a noise in the noisy environment, immediately before the input speech is input; detecting an input speech feature parameter X representing a cepstrum parameter of the input speech in the noisy environment; calculating a modified clean speech feature parameter Y according to a following equation:

Y=k·S+(1-k)·N (0<k≦1),

where the "k" is a predetermined value corresponding to a signal-to-noise ratio in the noise environment; comparing the input speech feature parameter X with the modified clean speech feature parameter Y; and recognizing the input speech by repeatedly carrying out the calculating process and the comparing process with respect to the plurality of clean speech models.

Claims

1. A speech recognition method of recognizing an input speech in a noisy environment by using a plurality of clean speech models, each of said clean speech models having a clean speech feature parameter S representing a cepstrum parameter of a clean speech thereof, said speech recognition method comprising the processes of:

detecting a noise feature parameter N representing a cepstrum parameter of a noise in said noisy environment, immediately before said input speech is input;

detecting an input speech feature parameter X representing a cepstrum parameter of said input speech in said noisy environment;

calculating a modified clean speech feature parameter Y according to a following equation;

space="preserve" listing-type="equation">Y=k·

S+(1-k)·

N (0<

k≦

1),

where said "k" is a predetermined value corresponding to a signal-to-noise ratio in said noise environment;

comparing said input speech feature parameter X with said modified clean speech feature parameter Y; and

recognizing said input speech by repeatedly carrying out said calculating process and said comparing process, respectively, with respect to said plurality of clean speech models.

2. A speech recognition method according to claim 1, wherein said plurality of clean speech models are continuous hidden Morkav models each including parameters indicating state transition probabilities, mean vectors and variances, respectively.

3. A speech recognition method of recognizing an input speech in a noisy environment by using a plurality of clean speech models, each of said clean speech models having a clean speech feature parameter S representing a cepstrum parameter of a clean speech thereof, said speech recognition method comprising the processes of:

detecting a noise feature parameter N representing a cepstrum parameter of a noise in said noisy environment, immediately before said input speech is input;

detecting an input speech feature parameter X representing a cepstrum parameter of said input speech in said noisy environment;

calculating a modified input speech feature parameter Z according to a following equation;

space="preserve" listing-type="equation">Z={X-(1-k)·

N}/k (0<

k≦

1),

where said "k" is a predetermined value corresponding to a signal-to-noise ratio in said noise environment;

comparing said modified input speech feature parameter Z with said clean speech feature parameter S; and

recognizing said input speech by repeatedly carrying out said calculating process and said comparing process, respectively, with respect to said plurality of clean speech models.

4. A speech recognition method according to claim 3, wherein said plurality of clean speech models are continuous hidden Morkav models each including parameters indicating state transition probabilities, mean vectors and variances, respectively.

5. A speech recognition apparatus for recognizing an input speech in a noisy environment by using a plurality of clean speech models, each of said clean speech models having a clean speech feature parameter S representing a cepstrum parameter of a clean speech thereof, said speech recognition apparatus comprising:

a first detecting device for detecting a noise feature parameter N representing a cepstrum parameter of a noise in said noisy environment, immediately before said input speech is input;

a second detecting device for detecting an input speech feature parameter X representing a cepstrum parameter of said input speech in said noisy environment;

a calculating device for calculating a modified clean speech feature parameter Y according to a following equation;

space="preserve" listing-type="equation">Y=k·

S+(1-k)·

N (0<

k≦

1),

where said "k" is a predetermined value corresponding to a signal-to-noise ratio in said noise environment; and

a comparing device for comparing said input speech feature parameter X with said modified clean speech feature parameter Y in order to recognize said input speech.

6. A speech recognition apparatus according to claim 5, wherein said plurality of clean speech models are continuous hidden Morkav models each including parameters indicating state transition probabilities, mean vectors and variances, respectively.

7. A speech recognition apparatus for recognizing an input speech in a noisy environment by using a plurality of clean speech models, each of said clean speech models having a clean speech feature parameter S representing a cepstrum parameter of a clean speech thereof, said speech recognition apparatus comprising:

a first detecting device for detecting a noise feature parameter N representing a cepstrum parameter of a noise in said noisy environment, immediately before said input speech is input;

a second detecting device for detecting an input speech feature parameter X representing a cepstrum parameter of said input speech in said noisy environment;

a calculating device for calculating a modified input speech feature parameter Z according to a following equation;

space="preserve" listing-type="equation">Z={X-(1-k)·

N}/k (0<

k≦

1),

where said "k" is a predetermined value corresponding to a signal-to-noise ratio in said noise environment; and

a comparing device for comparing said modified input speech feature parameter Z with said clean speech feature parameter S in order to recognize said input speech.

8. A speech recognition apparatus according to claim 7, wherein said plurality of clean speech models are continuous hidden Morkav models each including parameters indicating state transition probabilities, mean vectors and variances, respectively.

9. A program storage device readable by a speech recognition apparatus tangibly embodying a program of instruction executable by said speech recognition apparatus to perform method processes for recognizing an input speech in a noisy environment by using a plurality of clean speech models, each of said clean speech models having a clean speech feature parameter S representing a cepstrum parameter of a clean speech thereof, said method processes comprising:

detecting a noise feature parameter N representing a cepstrum parameter of a noise in said noisy environment, immediately before said input speech is input;

detecting an input speech feature parameter X representing a cepstrum parameter of said input speech in said noisy environment;

calculating a modified clean speech feature parameter Y according to a following equation;

space="preserve" listing-type="equation">Y=k·

S+(1-k)·

N (0<

k≦

1),

where said "k" is a predetermined value corresponding to a signal-to-noise ratio in said noise environment;

comparing said input speech feature parameter X with said modified clean speech feature parameter Y; and

recognizing said input speech by repeatedly carrying out said calculating process and said comparing process, respectively, with respect to said plurality of clean speech models.

10. A program storage device according to claim 9, wherein said plurality of clean speech models are continuous hidden Morkav models each including parameters indicating state transition probabilities, mean vectors and variances, respectively.

11. A program storage device readable by a speech recognition apparatus tangibly embodying a program of instruction executable by said speech recognition apparatus to perform method processes for recognizing an input speech in a noisy environment by using a plurality of clean speech models, each of said clean speech models having a clean speech feature parameter S representing a cepstrum parameter of a clean speech thereof, said method processes comprising:

detecting a noise feature parameter N representing a cepstrum parameter of a noise in said noisy environment, immediately before said input speech is input;

detecting an input speech feature parameter X representing a cepstrum parameter of said input speech in said noisy environment;

calculating a modified input speech feature parameter Z according to a following equation;

space="preserve" listing-type="equation">Z={X-(1-k)·

N}/k (0<

k≦

1),

where said "k" is a predetermined value corresponding to a signal-to-noise ratio in said noise environment;

comparing said modified input speech feature parameter Z with said clean speech feature parameter S; and

recognizing said input speech by repeatedly carrying out said calculating process and said comparing process, respectively, with respect to said plurality of clean speech models.

12. A program storage device according to claim 11, wherein said plurality of clean speech models are continuous hidden Morkav models each including parameters indicating state transition probabilities, mean vectors and variances, respectively.